We forecast how many votes Berlusconi and Veltroni's coalitions are going to obtain on April 13th and 14th. Results are obtained looking at on-line media visibility of the two candidates and parties. We bet that PdL will get 44.67% of the preferences and PD 38.37%. However, our results explanatory power does not depend on how close our predictions will be. Our data show already an interesting result: polls’ result tend to converge to our predictions.

Literature on relationships between mass media and voting behaviour suggests two points of view for interpretating a strong correlation between frequency of media appearence and election results. The agenda setting theory claims the media are able to affect vote choice⁽¹⁾. On the other hand, is also possible to invert cause and effect arguing that media adapt themselves to the opinion of media users, hencefore reinforcing existing opinions rather than changing them⁽²⁾. Our analysis assumes that both these effects are present in the relationship between appearance on the web and election results, in a two-way process that has been effectively named “reinforcing spirals”⁽³⁾.

The strong correlation between electoral results and visibility on web resources has already been prooved. We have applied a similar analysis to the PD's primary elections last October⁽⁴⁾ and close results were obtained by Grippa and Del Vecchio(5). We now show how the online visibility in the last two months can be use to find electoral polls convergence and (we bet) to forcast electoral results

The database: Bayes-Swarm is a research project that developed an engine to extract information from internet sources. So far once a day it visits 30 sources. In each source the number of appearances of every word is saved and stored in a database. All time series may be consulted at www.bayes-swarm.com. Moreover, Bayes-Swarm saves and stores all the entire monitored webpages. This allows “a posteriori” researches and controls (e.g. a “Time-Machine function”, see reported links for details).

The Bayesfor Prediction

We have considered web visibility from February the 4th to April the 8th, Popolo della Libertà visibility is defined as the sum of the word “Berlusconi” and the word “PdL” (similarly “Veltroni” and “PD” for Partito Democatico). The two months where devided into 23 intervals of 3 days each. The visibility trends are the blue dashed line in the figure. The value plotted are the percentage of visibility in each 3 days period (the other considered coalitions are: La Sinistra l’Arcobaleno, La Destra, UdC, and PS).

The solid horizontal lines show the average visibility percentage in the two months: 44.67% and 38.37% respectively. We believe that this avarage will be very close to the electoral results. We may be wrong however, a second powerfull result can be shown: the convergence of the polls toward our preditction.

We considered 61 polls published from February the 4th and March the 27th. The period was then devided in 19 intervals of 3 days and the average polls results for PdL and PD were calculated (plotted in red). As shown in the figure, PD percentage polls series approaches the average visibility from below while PdL approaches our estimate from above and below.

In both cases web visibility of coalition and leader’ names seem to correctly predict electoral results.

You may download the databases we used at the page: www.bayesfor.eu/wiki/bayes-swarm/dati Find more information about Bayes-Swarm at www.bayes-swarm.com Contact us: info@bayesfor.eu

References:

• McCombs M. E. (2002) “The Agenda-Setting Role of the Mass Media in the Shaping of Public Opinion”, Mass Media Economics 2002 Conference, LSE, London, UK.
• Gentzkow M. and Shapiro J. M. (2005) “Media Bias and Reputation”, NBER Working Paper No. 11664.
• Slater M. D. (2007) “Reinforcing Spirals: The Mutual Influence of Media Selectivity and Media Effects and Their Impact on Individual Behavior and Social Identity”, Communication Theory, N. 58.
• Associazione Bayesfor (2008) “Visibilità mediatica dei candidati alle primarie del Partito Democratico. Analisi con i dati di Bayes-Swarm”, Il Politico, N. 217.
• Grippa F. and Del Vecchio P. (2008), “Take me to your Leader: Predicting Political Leadership using Social Network Metrics”, accepted for presentation at Sunbelt - International Social Network Conference, Florida, USA.

List of the polls included in the analysis